How we score bang for your buck
Cheap and barely adequate is not a value pick. A $799 Mini with 256GB storage will always look “efficient” if you divide capability by price. We do not do that.
- Hard filters. Body style, new vs Apple Certified Refurbished, in-stock, plus optional minimums for RAM, CPU cores, GPU cores, and GPU Neural Accelerators. We drop 256GB drives unless you only asked for Office.
- Fit. How close the machine is to the RAM, GPU, and storage we think that job wants. Extra RAM helps, with a cap, so a 128GB Studio cannot steal every ranking.
- Value.
score = 100 × fit² / √price. Doubling fit matters more than shaving a few dollars.
Each ticket also shows price per GB of RAM, per CPU core, per GPU core, and per GPU Neural Accelerator. We do not price the 16-core Apple Neural Engine. Apple’s MLX on M5 write-up is the reason: time-to-first-token is compute-bound on GPU Neural Accelerators (one per M5 GPU core, up to ~4× vs M4). Later tokens are memory-bandwidth-bound (M5 153 GB/s vs M4 120 GB/s). Which model fits is unified memory. ANE core count does not show up in those numbers, so it is not a floor and not a unit price.
Headless machines (Mini, Studio) get a $299 display added to the compared price unless you said you already have a monitor. Out-of-stock certified refurb is a note, not an offer. Prices are a curated snapshot, not a live scrape of Apple or Amazon. Each Studio SKU is a real Apple configure URL (chip + CPU + GPU + RAM + storage). RAM is not an independent upgrade: on Max chips a bigger memory option is a different CPU/GPU die. npm test fails if a hand-checked Apple list price in data/apple-list-checks.json disagrees with the catalog, or if a SKU pairs RAM with a die Apple does not sell.